Re: LANG: Text in VRML

Linas Vepstas (linas@innerdoor.austin.ibm.com)
Wed, 22 Mar 1995 18:33:53 -0600


Hi,

Whoops! I am trying hard not to flame in the note below ...

> On Mon, 20 Mar 1995, Linas Vepstas wrote:
>
> > Hi,
> >
> > I just found this stale note in my reader:
> >
> > > As far as "free" texture mapping, take a look at Doom running on
> > > a PC sometime. Software renderers are much better at doing that kind of
> > > stuff than current SGI hardware. I've got an indy and a pentium on my
> > > desk. Using commercially available software renderers, the pentium kills
> > > the indy on texture performance (ie: it can do it without a problem), and
> > > is pretty close to polygon performance. This is without specialized 3D
> > > hardware. For the most part I think that a lot of the specialized 3D
> > > hardware will just make OpenGL tolerable on a PC, because right now it's
> > > horrible.
> >
> > I am very concerned about the tone that this note is taking. This is an
> > apples-to-oranges comparison. OpenGL texture mapping is "general",
> > where the texture can be strecthed, shrank, oriented in whatever way on
> > whatever shape, while DOOM only works on "traps" (trapezoids -- quads
> > with horizontal or vertical edges (or degenerate triangles), with textures
> > aligned to match. The quality of the image is far different -- get close
> > to a wall in DOOM, and tell me that's high-quality texture mapping. Then
> > there's the age-old argument about sorting polygons vs. depth-buffering.
> > It's just a different class of features/functions, and the resulting
> > performance is (very) different.
>
> I am very concerned with the tone of your reply. I think the orginal

I percieved the original note as OpenGL-bashing. I was attempting to
defend OpenGL, describe why its slower than doom, and to characterize
what the traditional OpenGL marketplace is like, and what it's customers
look for. I did not mean to put down DOOM. DOOM is excellent.

> poster was referring to Doom to make a point about differences in
> performance, not exact comparisons. Doom is obviously a toy. His point is
> that even for a toy, it comes close, and it is a very successful product.
> Maybe we should think about all those PCs out there that DON'T have extra

Yes. Absolutely.

> special hardware for all this intricate rendering. The average consumer
> isn't going to want to buy a hulking SGI box just to do virtual
> walkthroughs of the local shopping mall.

Agreed.

> > The point is that no serious CAD/CAM vendor would even think of using DOOM
> > for thier engineering drawings -- its too broke, and too feature-limitied
> > in too many ways. OpenGL was/is designed to be a very general-purpose,
> > do-it-all, easy-to-use, h/w-acclerated 3D API. DOOM was designed to be
> > a cut-corners, never-mind-the-hacks, who-cares-if-looks-wiggy-if-its-fast
> > game application. It's kind of a mistake to point at DOOM and say, "do
> > that", and "ohh, do that in a general-purpose way".
>
> CAD/CAM? So 3D is reserved only for high minded Visual Engineers doing
> animated 3D renderings of jet engines and the Space Shuttle?

Never said that. Don't put words in my mouth.

> Get this: Everything that we do eventually has to have a purpose, or it's
> worthless. 3D is going to be a big part of our lives, sooner or later.
> Sooner, if we are allowed to free associate and invent, rather than make
> revservations, and limitations. I see Doom at one end of a scale. It's the
> base line. What a CAD/CAM vendor will produce is at the other end. In the
> middle is a Bell curve, around the peak of which are all the products
> about to be invented and marketed. Go too high on the curve, and your
> limited market won't allow you to make enough money to cover costs. Go too
> low, and you're one-upped. Stay in the middle, and you sell good products
> to large numbers of people, whose buying patterns then suggest other
> advances. It's a process. Look past Doom, to the next step. Think! Don't
> throw wet blankets.

I thought, and now I'll throw a real wet blanket:

Is it obvious to everyone on this list that the current vrml, as spec'ed,
can be made to run fast on a pentium/power-pc class machine?

It's NOT obvious to me. Clearly, VRML, in the form of Iris Inventor,
runs fine on SGI boxes (and I've seen betas on other platforms). But
in all cases, (SGI & betas) Inventor is built on OpenGL, which had
just been flamed for poor performance. So ... what are we saying, again?

Is someone implying that vrml is being ported to a doom-like code base
for performance? If so, Great! Yes! Do It! Lets discuss what we like/hate
about the current VRML w.r.t. some non-OpenGL rendering engine! Lets fix
the vrml spec so that it runs fast, lean, mean on one of these other API's!

To the best of my knowledge, no one anywhere has yet indicated, implied,
insinuated, announced, anything about a/an/some/any implementation of
VRML that DIDN'T run on top of OpenGL. So if you don't like the
performance of OpenGL, just wait till you see VRML. So, lesse, where did
that wet blanket go?

> It's about time 3D came out of the lab. Doom is that first big step. Give
> it credit for sparking the minds of all those little game crazy
> teenagers, some of whom will some day join our ranks as top notch
> computer scientists and engineers, building what THEY envision.

Jeez, I did NOT put down DOOM !

--linas